AgentMesh — AI Governance

Intercept AI prompts, enforce token quotas, cache responses. Works with Claude, ChatGPT, Gemini.

AgentMesh is an open-source governance proxy that sits between your browser and every AI tool you use. HOW IT WORKS The extension intercepts prompts typed into ChatGPT, Claude.ai, and Gemini before they reach the LLM. It checks a local semantic cache — if the same question was asked before (even worded differently), it returns the cached answer instantly at zero cost. WHAT IT DOES • Semantic cache — catches rephrased prompts at 0.70 cosine similarity (85% hit rate) • Token quota enforcement — shows your team's budget usage on every prompt • Governance overlay — displays cache hit/miss and quota status before each submission • Cost tracker — popup shows total tokens saved and cost saved across your session • Stats persist — cumulative savings survive browser restarts REAL NUMBERS 20 requests, 5 topic clusters, 4 phrasings each: → 85% cache hit rate → 75% cost reduction → 3 misses (cold-start only) REQUIRES A locally running AgentMesh proxy (free, open source): pip install agentmesh-proxy sentence-transformers agentmesh serve --port 8080 --demo OPEN SOURCE Apache 2.0 — github.com/anilatambharii/agentmesh